Replication Consistency Research In Parallel File System

With the development of information technology, people request high capacity andhigh performance on information storage, so cloud storage arises at the historic moment.In the cloud storage service, people focus on the performance and reliability of the cloudstorage. Cloud storage usually reflect as the distributed file system, there will need toimprove the performance and reliability of the distributed file system. Like the RAIDtechnology into the distributed file system, the introduction of replication can be a verygood manner to improve the performance and reliability of the system. How to establish aproper replication management mechanism to reduce the losses of the performance whatmaintaining replication consistency brings to, has becomes a hot topic in the research ofthe distributed file system.Google’s GFS, Amazon’s Dynamo and so on have proposed several solutions forreplication consistency, but they only focus on one part of these issues. With thedevelopment of the replication and load balance in the distributed file system, this papermake full use of load information and the heat of the file access information in storagenodes, put forward the replication consistency maintenance algorithm based on the heatinformation and the size of data. The algorithm gains the heat queue from the storagenodes terminally, and compares it with the update queue, if both have no intersection, thenthe updating can become slow, conversely, select the renewal which is heater and smallerto update first, then update the renewal which is not heat and larger. Usually maintainingthe replication consistency would increase the load of the metadata server. When theupdate affects the function of normal service, the maintenance of the replicationconsistency would switch to client which can reduce the load of the metadata server.The experiments show that, the strategy of the maintenance of replication consistencybased on the heat and size and dynamic consistency switching, realize the effectivemaintenance of the replication consistency and greatly reduce the costs in the maintenanceof the replication consistency. In the case which the renewal takes up more than20%of allrequest and several replication numbers, dynamic consistency have some improvementcompared with the traditional replication consistency maintenance strategy on thesystem’s response time.